Predicting Drug-Target Interactions Based on Small Positive Samples.

Area of Science:

• Computational chemistry and bioinformatics
• Drug discovery and development

Background:

• Drug discovery aims to identify compounds interacting with target proteins.
• Drug-target interactions (DTIs) are crucial for medication efficacy and are not random.
• Discovering patterns in DTIs can accelerate the identification of new drugs.

Purpose of the Study:

• To address limitations of existing DTI prediction methods, such as inability to directly consider biochemical features and computational expense.
• To develop a novel computational approach for predicting drug-target interactions (DTIs) that overcomes current challenges.
• To improve the accuracy and efficiency of DTI prediction by leveraging known positive interactions.

Main Methods:

• Introduced ODT (one-class drug target interaction prediction), a novel approach for DTI prediction.
• Transformed the DTI network into a representation based on structural properties of drugs and protein sequences.
• Employed a one-class classification algorithm using only known positive interactions to build a predictive model.

Main Results:

• ODT demonstrated superior prediction accuracy compared to state-of-the-art methods on Gold standard datasets.
• The method achieved the best AUROC scores on GPCRs and Ion channels datasets.
• ODT's performance indicates its potential for accurate DTI prediction.

Conclusions:

• ODT is a potentially valuable tool for predicting drug-target interactions.
• Predicting DTIs from known interactions is a promising strategy to overcome issues with unreliable negative samples and high computational costs.
• The approach offers a more efficient and reliable method for drug discovery.